Correlated double sampling (CDS) is a technique used in image sensors for reducing pixel noise. Usually implemented in the readout, the CDS includes sampling of both a reset voltage and a signal voltage. The two voltages are then subtracted in analog domain (analog CDS) or first converted by an analog-to-digital converter and then subtracted in the digital domain (digital CDS). As a result, such noise reduction technique using sampling of two voltages implies that acquiring a pixel value is associated with a double reading of a pixel voltage, such that a speed of acquiring the pixel value is reduced by a factor 2. This may be very critical for high speed sensors.
In order to avoid the reduction in speed, CDS may be implemented in-pixel and the difference between the signal and the reset voltage is provided at output of the pixel in a single reading of a pixel voltage.
When the CDS is done at column level (analog CDS) or after the analog-to-digital conversion (digital CDS), a fixed pattern noise (FPN) due to the pixel array is removed. However, when in-pixel CDS is used, there may be a remaining FPN in read-out of pixel signals which is not removed. Therefore, it would be desired to further improve FPN handling in image sensors using in-pixel CDS.